Narrative is one of the fundamental cognitive tools that we use to understand the world around us. When interacting with other humans we rely on a shared knowledge of narrative structure, but in order to enable this kind of communication with digital artifacts we must first formalize these narrative conventions. Narratology, computer science, and cognitive science have formed a symbiotic relationship around this endeavor to create computational models of narrative. These models provide us a deeper understanding of story structure and will enable us to create a fundamentally new kind of interactive narrative experience in which the author, the audience, and the machine all participate in the story composition process.
This document presents a computational model of narrative conflict, its empirical evaluation, and its deployment in an interactive narrative experience. Narratologists have described conflict in terms of the difficulties that an intelligent agent encounters while executing a plan to achieve a gol.. This definition is inherently plan-based, and has been integrated into an existing model of narrative based on the data structures and algorithms of artificial intelligence planning|the process of constructing a sequence of actions to achieve a goal. The conflict Partial Order Causal Link (or CPOCL) model of narrative represents the events of a story along with their causal structure and temporal constraints. It extends previous models by representing non-executed actions which describe how an agent intended to complete its plans even if those plans failed, thus enabling an explicit representation of thwarted plans and conflict. The model also includes seven dimensions which can distinguish one conflict from another and provide authors with greater control over story generation: participants, topic, duration, balance, directness, intensity, and resolution.
One valuable aspect of plan-based models is that they can be generated and modified automatically. Two story creation methods are discussed: the plan-space CPOCL algorithm that works directly with the rich CPOCL knowledge representation and the state-space Glaive algorithm which is significantly faster. Glaive achieves its speedup by incorporating research from fast forward-chaining state-space heuristic search planning and by using the constraints that a valid narrative plan must obey to calculate a more accurate heuristic. Glaive is fast enough to solve certain non-trivial narrative planning problems in real time.
This computational model of narrative conflict has been evaluated in a series of empirical experiments. The first validates the three discrete dimensions of conflict: participants, topic, and duration. It demonstrates that a human audience recognizes thwarted plans in static text stories in the same places that the CPOCL model defines them to exist. The second experiment validates the four continuous dimensions|balance, directness, intensity, and resolution|by showing that a human audience ranks static text stories in the same order defined by the formulas for those dimensions.
The final experiment is an evaluation of an interactive narrative video game called The Best Laid Plans, which uses Glaive to generate a story at run time from atomic actions and without recourse to pre-scripted behaviors or story fragments. In this game, the player first acts out a plan to achieve a goal and then Glaive coordinates all the non-player characters in the game to thwart the player's plan. The game is evaluated relative to two other versions: a control in which the other characters do nothing and a scripted version in which the other characters are controlled by programs written by a human author. Players recognize intentionality and conflict in the stories Glaive produces more so than in the control and comparably to the human scripted version.
In summary, this document describes how a narratological definition of conflict as thwarted plans has been operationalized in plan data structures and incorporated into a narrative planning algorithm. The knowledge representation is rich enough that a human audience recognizes thwarted plans where the model defines them to exist. The algorithm is fast enough to be used in a real time interactive context for certain non-trivial story domains. This work represents one small advancement toward understanding human storytelling and leveraging that understanding in interactive systems.
|Advisor:||Young, R. Michael|
|School:||North Carolina State University|
|School Location:||United States -- North Carolina|
|Source:||DAI-B 76/05(E), Dissertation Abstracts International|
|Keywords:||Human storytelling, Narrative conflict, Planning|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
The supplemental file or files you are about to download were provided to ProQuest by the author as part of a
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
Copyright of the original materials contained in the supplemental file is retained by the author and your access to the
supplemental files is subject to the ProQuest Terms and Conditions of use.
Depending on the size of the file(s) you are downloading, the system may take some time to download them. Please be